Detection and Attribution of Climate Change... in Indices of Extremes? Reiner Schnur Max Planck Institute for Meteorology MPI-M Workshop Climate Change Scenarios and Their Use for Impact Studies September 29-30, 2005
Outline Detection and Attribution IPCC AR4 Data Archive Optimal Bayesian D&A for Extremes in Europe
Detection and Attribution Conventional Multi-pattern linear regression between observations and a few (-3) spatio-temporal patterns of model response to different forcings. Optimized to enhance signal-to-noise ratio
Optimal Bayesian Detection & Attribution posterior odds = likelihood ratio * prior odds Explicit use of natural variability AND model uncertainties Optimal filtering maximizes impact of evidence on prior probability of anthropogenic forcing hypothesis Less severe dimensional reduction than in conventional D&A No distinction between detection and attribution Allows inclusion of prior information Schnur and Hasselmann, 2005
Detection and IPCC AR4 Data Archive Large ensemble from 23 models Relatively long control simulations (Picntrl) Problems: Only 20C3M, no historic experiments with different forcings (needed to use spatio-temporal signals) Needed for attribution: ALL, NAT, ALL-GHG, ALL-GS, etc. (alternatively ALL, NAT, GHG, GS, etc.) 20C3M: some models don't include natural forcings (solar and volcanic)
Detection and Attribution of Extremes Increased interest in (regional) changes in extremes because more relevant to people largest impact on societies but more difficult than global because larger variability on small scales resolution of climate models data availability
Extremes Indices Longer observations and shorter return periods give higher detection probability. The indices are therefore based on daily data with return periods of months rather than years (i.e. no 00-yr floods). Total number of frost days (fd) Intra-annual extreme temperature range (etr) Growing season length (gsl) Heat wave duration index (hwdi) Fraction of time with Tmin > 90th percentile of Tmin for baseperiod (tn90) Number of days with precipitation > 0 mm/day (r0) Maximum number of consecutive dry days (cdd) Maximum 5-day precipitation total (r5d) Simple daily intensity index (sdii) Fraction of annual total precipitation due to evens > 95th percentile in baseperiod (r95t) Frisch et al, 2002
Observations for Extremes Indices Need daily observations for temperature and precipitation (global, regional) ECA&D European Climate Assessment and Dataset (stations) CCL/CLIVAR Working Group on Climate Change Detection Kiktev et al, 2003 (gridded indices) Schnur (gridded tmin/tmax/precip 979-2003) (Nijssen et al, 200)
Models (Extreme Indices) Model Picntrl 20C3M SRESAB CNRM-CM3, France GFDL-CM2.0, USA GFDL-CM2., USA 3 3 INM-CM3.0, Russia MICROC3.2 (hires), Japan MICROC3.2 (midres), Japan MRI-CGCM2.3.2, Japan 3 5 3 5 NCAR-CCSM3, USA NCAR-PCM, USA 0 (?) 0 7 (2780 yr) 4 22 4 8 Problem: many scenario simulations contain only one member (model uncertainty vs. internal variability)
Detection and Attribution of Extremes Data Preparation D&A incredients: Expected climate change signal mean of all SRESAB of 'mean(207-90)-mean(20-30)' inter-model covariance: model uncertainty + natural variability Observations trend from 979-2003 Natural variability variability of 25-yr trends of concatenated control runs All data aggregated to a 0 x0 grid covering Europe.
Consecutive Dry Days Heat Wave Duration Index Fraction P from > 95%tile SimpleDaily Intensity Index Extremes Indices Variability of 25-yr trends from all controls (Europe)
Extremes Indices 20C3M Simulations and Observations Heat wave duration index Differences between models but trend is captured quite well.
Baysian Detection of Extremes (Europe) Results Based on the Bayes Factors (likelihood ratios) for the anthropogenic forcing hypothesis given by SRESAB scenario and the natural variability hypothesis, the following indices can be detected: cdd sdii r95t hwdi Analysis A yes yes yes yes Analysis B no yes no yes
Extreme Indices Signals SRESAB Maximum number of consecutive dry days change/yr
Extreme Indices Signals SRESAB Heat wave duration index change/yr
Extreme Indices Signals SRESAB Fraction of annual total precipitation due to evens > 95th percentile in baseperiod change/yr
Extreme Indices Signals SRESAB Simple daily intensity index change/yr
Discussion Preliminary analysis suggests that regional detection and attribution for indices of extremes seems to be possible but more work needs to be done. IPCC AR4 data archive is very useful but also limited (forcing mechanisms, availability of indices). Need to analyze other regions get better (longer) observations fully compute extremes indices for MPI-M model and submit to data archive
The End